A Novel Binary Particle Swarm Optimization Algorithm for Feature Selection

Feature selection is an important task in machine learning and binary feature selection is a challenging problem due to the large search space. Swarm intelligent optimization algorithms have received great attention in solving numerous optimization problems. Among them, the particle swarm optimization (PSO) algorithm has successfully solved many optimization problems. However, there are still some difficulties with PSO in discrete or binary versions that lead to slow convergence. To solve these problems, a novel binary particle swarm optimization (NBPSO) algorithm is proposed in this paper. NBPSO introduces a dynamic weight search strategy to meet the different requirements for solution performance in different stages. Meanwhile, it combines with Lévy flight search strategy to reduce the possibility of falling into local optimal solution. In addition, referring to the idea of genetic algorithm, we introduce a population improvement method of variation mechanism and high-quality individual replication to balance the exploration and development ability of NBPSO algorithm. In order to verify the effectiveness of the proposed method and explore its performance on the feature selection problem, we use several benchmark functions to test and conduct comparative experiments on six classical datasets. The results show that the overall performance of the proposed algorithm is better than other comparison algorithms.